Matthias E.

I developped a chatbot with Shrt-Term-Memory and Long-Terme-Memory management and RAG time weighted vectorization. This chatbot is working with differents LLM models you can choose and Embedding models also. The chatbot is working on vectore database like FAISS or Pinecone.

I use Langchain to wrapp everything. The chatbot has a personality thanks to prompt engineering and example like few shot learning.

Integrated Retrieval and Summarization System for Compapny Web Content:

Led the development of an advanced system combining Information Retrieval and Text Summarization for Comapny web content. Utilizing Web API, Textual Context Similarity, and Large Language Models (LLMs), the system was specifically designed to extract and condense long texts It efficiently summarized key information like company services, product lists, executive teams, and contact details. This framework not only streamlined the process of integrating new features into our products but also ensured the extraction of relevant and specific information, significantly aiding in our product development and market analysis efforts.

Synonyms Matcher for Data Normalization:

Developed a synonyms matcher using Recurrent Neural Networks (RNN) to normalize company solution names. This innovation reduced data conflicts by 30%, significantly improving data quality and consistency in our processes.

Team Diversity Score Development:

Created a unique Team Diversity Score for new product features by training a name ethnicity classification model. This tool assesses diversity within groups, contributing to our commitment to inclusivity and varied perspectives in product development.

Topic Modeling for Data Relevancy and Insights:

Built a Topic Modeling Model from scratch, inspired by Entrio.io’s taxonomy. This model is key in assessing new data insights and verifying the relevancy of incoming data, ensuring that our data-driven strategies are always aligned with our core objectives and values.

Developed pipeline for storing, cleaning and engineering data to enrich databases.

Implemented a density-based algorithm to assess customer’s needs.

Build statistical tools to assist annotation, recognize patterns and features selection.

Developed classification model to automate 50+IT tasks, resulting in a 25% reduction in average tasks number.

Used technologies such as BERT, TF-IDF, kNN, UMAP, Decision Trees, SMOTE, and Azure Storage and Azure ML Studio, MongoDB.